Abstract:
Mortality rate has traditionally been used as an essential health indicator for assessing population well-being and has consistently gained prominence in the Millennium Development Goals (MDGs) and Sustainable Development Goals (SDGs). Over the past decades, new models for mortality modeling have been developed. A pioneering and most influential model among others is the Lee-Carter model developed by Lee and Carter in 1992. Since its development, the model has undergone extensive review, and other variants with various structures suggested. While the model's effectiveness has been evaluated in various scenarios, its capacity to precisely handle non-linearity and long-term dependence in the mortality index over time has not been thoroughly examined. The Lee-Carter (LC) model's conventional approach exhibits evident limitations in predicting future mortality patterns. Therefore, this study focused on using a Gated recurrent unit integrated LC model to correct some of the shortfalls of the traditional LC model. Simulated data from Weibull and Gompertz distributions were used. Using the performance metrics (RMSE, MAE and MAPE) LC-GRU model showed promising results in handling the non-linearity and long-term dependence inherent in mortality data.
Description:
In actuarial science and demography, mortality predictions have a long history. Mortality estimates are utilized by actuaries for pension annuities, life insurance reserves, cash flow forecasts, and population projections. Government agencies use mortality forecasts to help them make decisions about funding for retirement income, social security, and healthcare (Oeppen and Vaupel, 2002). Having a clear understanding of future mortality and longevity risk is crucial for individuals in actuarial science and government authorities alike. While advancements in longevity lead to improved well-being and longer productive lives for many individuals, they also pose challenges for pension systems.
